Manas Documentation

Building a Research Flow

Learn how to create a multi-node research flow that can research, analyze, and summarize topics

Intermediate 30 minutes

Building a Research Flow

This tutorial demonstrates how to build a multi-node flow for researching topics, analyzing findings, and generating reports.

Overview

We’ll create a flow with three specialized nodes:

  • A researcher node that gathers information
  • An analyst node that identifies patterns and insights
  • A writer node that produces a final report

Prerequisites

1

pip install "manas-ai[openai]"  # Or your preferred provider

Implementation

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77

import os
from core import Flow, LLM
from manas_ai.nodes import QANode, RAGConfig

# Initialize LLM (using OpenAI as example)
model = LLM.from_provider(
    "openai",
    model_name="gpt-4",
    api_key=os.environ.get("OPENAI_API_KEY")
)

# Create nodes with specific roles
researcher = QANode(
    name="researcher",
    llm=model,
    config=RAGConfig(
        system_prompt=(
            "You are an expert researcher. Investigate the given topic thoroughly "
            "and provide comprehensive, factual information. Include relevant "
            "details, statistics, and context."
        ),
        use_rag=True  # Enable RAG for knowledge enhancement
    )
)

analyst = QANode(
    name="analyst",
    llm=model,
    config=RAGConfig(
        system_prompt=(
            "You are a data analyst specializing in finding patterns and insights. "
            "Analyze the research provided and identify key trends, implications, "
            "and notable findings. Be objective and thorough in your analysis."
        )
    )
)

writer = QANode(
    name="writer",
    llm=model,
    config=RAGConfig(
        system_prompt=(
            "You are a technical writer. Create a well-structured report based on "
            "the analysis provided. Use clear headings, concise language, and "
            "highlight key points. Format your response using Markdown."
        )
    )
)

# Create and configure the flow
flow = Flow()

# Add nodes
flow.add_node(researcher)
flow.add_node(analyst)
flow.add_node(writer)

# Connect nodes in sequence
flow.add_edge(researcher, analyst)
flow.add_edge(analyst, writer)

# Optional: Add knowledge base for RAG
if researcher.rag_node:
    researcher.rag_node.add_documents([
        "path/to/research_papers/",
        "path/to/reference_docs/"
    ])

# Process a research query
topic = "Recent advances in quantum computing and their implications"
result = await flow.process({
    "prompt": topic,
    "max_sources": 5  # Limit number of sources to include
})

print("\nResearch Report:")
print(result["writer"]["response"])

Understanding the Flow

Node Communication

  1. The researcher node receives the initial query and:
    • Uses RAG to enhance its knowledge
    • Generates comprehensive research findings
    • Includes relevant sources and citations
  2. The analyst node:
    • Receives the research output
    • Identifies key patterns and insights
    • Structures the information logically
  3. The writer node:
    • Takes the analyzed information
    • Creates a well-formatted report
    • Uses Markdown for clear presentation

Customization Options

You can customize the flow by:

  • Adjusting system prompts for different writing styles
  • Configuring RAG parameters for knowledge retrieval
  • Adding more nodes for specific tasks (e.g., fact-checking)
  • Modifying the flow structure for different workflows

Error Handling

1
2
3
4
5
6
7

try:
    result = await flow.process({"prompt": topic})
except Exception as e:
    print(f"Error in flow execution: {e}")
finally:
    # Always clean up resources
    await flow.cleanup()

Complete Example

A full working example of this research flow is available in our examples repository: research_flow_example.py

Complete Example

Find the complete example in our GitHub repository: code research_flow_example.py

Related Documentation